Dry-type air-core reactor turn-to-turn fault identification method based on acquired data optimization

Abstract

The invention relates to the technical field of health management of dry-type air-core reactors, in particular to a dry-type air-core reactor turn-to-turn fault identification method based on acquired data optimization. The method comprises the steps of firstly obtaining a discharge feature vector of each event; further obtaining associated defect source cost dispersion according to the same-dimensional data difference of the discharge feature vectors between adjacent monitoring periods; further obtaining the structure entropy of the discharge system according to the distribution of the discharge feature vectors in each monitoring period; further constructing an insulation state diagnosis space according to the change of the structure entropy of the discharge system in the adjacent monitoring periods and the cost dispersion of the associated defect source; and finally, according to the distribution of diagnosis points in the insulation state diagnosis space, carrying out multi-stage state evaluation, capturing early-stage degradation information, realizing accurate prediction of the insulation degradation severity and the dominant mode, quantifying the abnormal evolution process, and providing a reliable basis for health management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dry-type air-core reactor health management, and particularly relates to a dry-type air-core reactor inter-turn fault identification method with optimized data collection. BACKGROUND

[0002] In a scene containing a large number of power electronic devices or frequent switching operations such as a converter station, the repetitive electric stress of a dry-type air-core reactor can cause space charge accumulation in the internal space of the insulating material such as epoxy resin, induce partial discharge and eventually form a conductive channel to cause an inter-turn short circuit. The existing identification method for such a fault mainly relies on capturing the changes in current, voltage and other power frequency electrical quantities caused by the inter-turn short circuit current after the fault occurs, such as a fault recorder and differential protection.

[0003] However, the existing method is essentially a post-response mechanism, and there is a lag in fault identification and health management. Secondly, the weak discharge signal energy is extremely low in the early stage of insulation degradation, and it is difficult to effectively extract. Finally, the existing technology usually regards insulation degradation as a single process, which cannot effectively distinguish different degradation paths, thereby causing blindness in operation and maintenance decision-making, which may cause accidents due to the delay in handling high-risk defects, or unnecessary shutdown maintenance for low-risk states. SUMMARY

[0004] In order to solve the technical problems that the existing technology is difficult to effectively extract early insulation degradation information under strong transient overvoltage interference, and cannot accurately predict faults and manage health, the purpose of the present application is to provide a dry-type air-core reactor inter-turn fault identification method with optimized data collection, and the technical solution adopted is as follows: Obtain a discharge feature vector of each event record of each preset monitoring period; According to the same dimension data difference of the discharge feature vector between each monitoring period and the adjacent monitoring period, a cross-period defect source correlation cost matrix is constructed, and an optimal allocation is performed to obtain a correlation defect source cost dispersion; according to the distribution of the discharge feature vector in each monitoring period, a discharge system structure entropy is obtained; according to the change of the discharge system structure entropy of each monitoring period compared with the last monitoring period, in combination with the correlation defect source cost dispersion, an insulation state diagnosis space is constructed; According to the distribution of the diagnosis points in the insulation state diagnosis space, a multi-level state evaluation is performed.

[0005] Further, the discharge feature vector includes: a discharge time centroid, discharge energy, spectral centroid and response waveform entropy of the voltage response sequence of each event.

[0006] Further, the acquisition method of the cross-period defect source correlation cost matrix includes: In each of the monitoring period and the adjacent last monitoring period, one event is extracted respectively to form an event pair; According to the data difference of each dimension of the two discharge feature vectors, a cost component of each dimension corresponding to the event pair is obtained; all the cost components of each dimension form a cost component matrix, and a cross-period defect source correlation cost matrix is obtained after linear fusion with a preset weight.

[0007] Further, the cost component acquisition method comprises: In the event pair, the difference between the discharge time centroid of the event in the monitoring period with the largest time sequence and the discharge time centroid of the event in the monitoring period with the smallest time sequence is taken as the cost component of the corresponding dimension; the absolute values of the differences of the two events in the discharge energy, the spectral centroid and the response waveform entropy dimensions are taken as the cost components of the corresponding dimensions respectively.

[0008] Further, the correlation defect source cost dispersion acquisition method comprises: Based on the Jonker-Volgenant algorithm, the cross-period defect source correlation cost matrix is optimally distributed to obtain a set of associated defect source costs, and according to the dispersion characteristics of the elements in the set of associated defect source costs, a correlation defect source cost dispersion is obtained.

[0009] Further, the discharge system structure entropy acquisition method comprises: The Euclidean distance between the discharge feature vectors of any two events in each monitoring period is obtained, and according to the dispersion characteristics of all the Euclidean distances, a period discharge dispersion is obtained; based on the period discharge dispersion as the kernel width, the Euclidean distances are mapped by a Gaussian kernel function to obtain a discharge event similarity matrix; The graph Laplacian matrix of the discharge event similarity matrix is obtained, and the eigenvalue decomposition is performed to obtain the Shannon entropy of the eigenvalue as the discharge system structure entropy.

[0010] Further, the insulation state diagnosis space construction method comprises: The difference between the discharge system structure entropy of each monitoring period and the discharge system structure entropy of the adjacent last monitoring period is taken as the horizontal coordinate of the diagnosis point of each monitoring period; the correlation defect source cost dispersion is taken as the vertical coordinate of the diagnosis point of each monitoring period; The diagnosis points of all the monitoring periods are used to construct an insulation state diagnosis space.

[0011] Further, the multi-level state evaluation method comprises: When the diagnosis point of the monitoring period is in the preset target quadrant of the insulation state diagnosis space, mark the corresponding diagnosis point as a state evaluation point, obtain the square of the distance from the state evaluation point to the origin as a degradation severity index, and obtain the angle of the state evaluation point in the preset target quadrant as a degradation mode angle. According to the distribution of the degradation severity index of the state evaluation point, combined with the degradation mode angle, multi-level state evaluation is performed.

[0012] Further, the method of performing multi-level state evaluation according to the distribution of the degradation severity index of the state evaluation point, combined with the degradation mode angle, comprises: Based on the degradation severity index of all the state evaluation points, if the degradation severity index of the state evaluation points in the current preset time domain neighborhood is less than the health threshold, it is determined that the system state is normal. If the degradation severity index of the state evaluation points in the current preset time domain neighborhood is less than the health threshold and greater than or equal to the health threshold at the same time, an attention warning is issued. If the degradation severity index of the state evaluation points in the current preset time domain neighborhood is greater than or equal to the health threshold, a degradation warning is issued.

[0013] Further, the method of issuing a degradation warning further comprises: When the degradation mode angle of the latest state evaluation point is in a preset low angle interval, it is determined that a single structural defect is rapidly developing. When the degradation mode angle of the latest state evaluation point is in a preset high angle interval, it is determined that multi-point diffuse degradation is occurring. When the degradation mode angle of the latest state evaluation point is in a preset medium angle interval, it is determined that a mature dominant defect is evolving.

[0014] The present application has the following beneficial effects: The application firstly acquires the discharge characteristic vector of each event, provides a data basis for subsequent system dynamics analysis; further, according to the same dimension data difference of the discharge characteristic vectors between adjacent monitoring periods, a cross-period defect source correlation cost matrix is constructed and solved to acquire the correlation defect source cost dispersion, which provides a reliable input for subsequent insulation state evaluation; further, according to the distribution of the discharge characteristic vectors in each monitoring period, the discharge system structure entropy is acquired, the dispersed single discharge characteristic vector can be converted into an overall order index through the discharge system structure entropy, and the transformation process from the disordered state dominated by random micro-discharge to the ordered state dominated by local defects of the insulation system can be directly reflected; further, the change of the discharge system structure entropy of adjacent monitoring periods is combined with the correlation defect source cost dispersion to construct an insulation state diagnosis space, which can simultaneously quantify the global state and local abnormal deviation of the system, thereby forming a two-dimensional diagnosis space to realize more comprehensive state diagnosis; finally, the multi-level state evaluation is performed according to the distribution of the diagnosis points in the insulation state diagnosis space. The defect source correlation cost dispersion is constructed through the cross-period discharge characteristic difference, the insulation state diagnosis space is formed in combination with the change of the discharge system structure entropy, and the multi-level state evaluation is performed based on the distribution of the diagnosis points, early deterioration information is captured, the precise prediction of the insulation deterioration severity and the dominant mode is realized, and the abnormal evolution process is quantified, thereby providing a reliable basis for health management. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, a brief introduction will be given to the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only show some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0016] Figure 1 A flowchart of a data acquisition optimized dry-type air-core reactor inter-turn fault identification method provided by an embodiment of the present application; Figure 2 A flowchart of a cross-period defect source correlation cost matrix acquisition method provided by an embodiment of the present application; Figure 3 A flowchart of a multi-level state evaluation method provided by an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined inventive purpose, the following describes in detail the specific implementation, structure, features and effects of a kind of dry-type air-core reactor inter-turn fault identification method for collecting data optimization according to the present application, combined with the preferred embodiments and the drawings. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0019] The specific scheme of the dry-type air-core reactor inter-turn fault identification method for collecting data optimization provided by the present application is described in detail below in combination with the drawings.

[0020] Please refer to Figure 1 , which shows a flowchart of a dry-type air-core reactor inter-turn fault identification method for collecting data optimization according to an embodiment of the present application, which specifically includes: Step S1: Obtain the discharge feature vector of each event record in each preset monitoring period.

[0021] Since the discharge energy of the dry-type air-core reactor is extremely low in the early stage of insulation degradation, the ultra-high frequency (UHF) signal generated will be completely overwhelmed by the transient overvoltage pulses and background electromagnetic noise generated by the switching operations of the switching devices in the converter station if a continuous collection method is used, resulting in that the target signal cannot be effectively identified.

[0022] Therefore, the embodiment of the present application adopts an event-driven synchronous collection mechanism to collect data with a clear external transient overvoltage impact as a time reference, aiming to improve the signal-to-noise ratio of the target signal and establish the causal relationship for subsequent analysis.

[0023] Specifically, first, a ultra-high frequency (UHF) sensor is deployed on or near the reactor body, and the effective working frequency band of the sensor covers the main energy range of the insulation discharge signal, for example, 300MHz to 1.5GHz. The sensor is connected to a high-speed data collection unit. The system continuously monitors the sequence of events (SOE) recorded by the station monitoring and data acquisition (SCADA) system, which records the operation of the bus disconnecting switch or circuit breaker and other devices. Each SOE record corresponds to an event.

[0024] When receiving an SOE record, the system takes the precise timestamp indicating the operation of the switching device in the record as the reference trigger time for this collection . In At the moment, the high-speed data acquisition unit is triggered immediately to have a sampling rate of no less than 2GSa / s The output of the UHF sensor is collected. The collection lasts for a preset window length capable of covering a single transient process completely (e.g. 200 nanoseconds).

[0025] The preset monitoring period is set to be 1 day.

[0026] It should be noted that the sampling frequency and window length of all events are uniform, and the installation and setting of various sensors are prior art, and the implementer can select the sensor model and adjust the setting, and can also adjust other window length and monitoring period settings, which will not be described in detail.

[0027] Each event collects a time-domain transient voltage response sequence, which contains multi-dimensional information related to the discharge occurrence time, energy, generation mechanism and propagation path. Direct analysis of high-dimensional waveform data is not only computationally complex, but also difficult to quantitatively compare between events, so the discharge feature vector is extracted to provide a data basis for subsequent system dynamics analysis.

[0028] Preferably, in an embodiment of the present application, the discharge feature vector includes: the discharge time centroid, discharge energy, spectral centroid and response waveform entropy of the voltage response sequence of each event.

[0029] Taking any voltage response sequence as an example, the acquisition process of the discharge feature vector includes: Considering that the deterioration of the insulating medium will change the local electric field distribution, affect the time required for the formation of the discharge channel, and cause the energy center of the discharge signal to shift relative to the trigger time, this shift is a key time sequence feature representing the degree of deterioration, so the discharge time centroid is obtained.

[0030] As an example, the sampling points in each voltage response sequence are respectively time-sequenced, and the square value of the data value of each sampling point in the voltage response sequence represents the instantaneous energy of each sampling point. The sequence number of each sampling point is weighted by the instantaneous energy, and the product of the sequence number and the instantaneous energy of each sampling point is summed up, and the total sum value is taken as the numerator, and the total sum value of the instantaneous energy is taken as the denominator, and the fractional ratio is taken as the sequence number corresponding to the time centroid, and multiplied by the reciprocal of the sampling period. The product is taken as the discharge time centroid of the corresponding event.

[0031] Considering that the severity of the discharge is directly reflected in the total energy of the electromagnetic wave, the quantification of the total energy of the signal is the basis for evaluating the severity of a single event, and the discharge energy is obtained.

[0032] As an example, the sum of the instantaneous energy of all sampling points is taken as the discharge energy of the corresponding event, where the unit of the discharge energy is proportional to the total energy of the entire sampling window, and since the relative change trend of the energy is usually concerned rather than the absolute value, the proportional coefficient can be ignored or normalized.

[0033] Considering that different discharge types (such as surface discharge and internal air gap discharge) have different charge acceleration and recombination processes, resulting in electromagnetic wave signals with different spectral characteristics, the center of gravity of the signal energy in the frequency domain is an important basis for distinguishing the discharge generation mechanism, and the spectral centroid is obtained.

[0034] As an example, the time domain sequence is transformed into the frequency domain by fast Fourier transform to obtain the power spectrum in the frequency domain, and then the frequency average value weighted by the power spectrum is obtained as the spectral centroid. The sum of the product of the frequency value and the power value of each frequency in the power spectrum is taken as the numerator, and the sum of the power values is taken as the denominator, and the fractional ratio is taken as the spectral centroid.

[0035] Considering that the discharge signal will undergo multiple reflections, refractions and superpositions when propagating in the complex winding structure of the reactor, the complexity of the waveform form finally received by the sensor indirectly reflects the complexity of the discharge source position and propagation path. In order to quantify this form feature, Shannon entropy is applied to calculate the uncertainty of the waveform, i.e. to obtain the response waveform entropy.

[0036] As an example, the sequence of instantaneous energy of the sampling points is normalized, i.e. each instantaneous energy is divided by the sum of the instantaneous energy, and a probability distribution sequence is obtained after sorting the normalized results, so that the response waveform entropy is calculated by applying the Shannon entropy formula.

[0037] The discharge time centroid, discharge energy, spectral centroid and response waveform entropy obtained can be constructed as the discharge feature vector of a single discharge event.

[0038] It should be noted that the analysis process of the voltage response sequence of each event is consistent, and only one example is described here; the Shannon entropy formula, fast Fourier transform and power spectrum acquisition are well-known techniques and will not be described again.

[0039] Step S2: According to the same dimension data difference of the discharge feature vector between each monitoring period and the adjacent monitoring period, a cross-period defect source association cost matrix is constructed, and the associated defect source cost dispersion is obtained by optimal allocation; according to the distribution of the discharge feature vector in each monitoring period, the discharge system structure entropy is obtained; according to the change of the discharge system structure entropy of each monitoring period compared with the last monitoring period, combined with the associated defect source cost dispersion, an insulation state diagnosis space is constructed.

[0040] The existing method analyzes the discharge activity in the insulation interior as a series of isolated and static events, ignoring the dynamic evolution information contained therein which can reveal the real state of the degradation process. The embodiment of the present application regards the insulation degradation process as a dynamic system jointly driven by individual defect evolution and system overall structure evolution. Therefore, this step quantifies the two mutually orthogonal evolution dimensions by analyzing the changes of the discharge characteristic vectors of adjacent monitoring periods, and finally provides a basis for subsequent fusion evaluation in a unified diagnostic space.

[0041] Firstly, the individual evolution path of the defect source is identified and quantified. Considering that the surrounding medium of a single micro defect (such as a micron-sized air gap in an epoxy resin casting) in the insulation interior will be gradually damaged under the repeated action of transient overvoltage, resulting in more serious local electric field distortion, which makes the subsequent discharge more easily triggered by the rising front of the voltage pulse. In order to identify and track the path representing the continuous evolution of the same defect source in time from the massive and discrete discharge events in the monitoring period, and quantify its evolution behavior, this step realizes it by constructing and solving a cross-period defect source correlation cost matrix C.

[0042] Therefore, according to the same dimension data difference between the discharge characteristic vectors of each monitoring period and the adjacent monitoring period, a cross-period defect source correlation cost matrix is constructed to capture the dynamic change trend of insulation degradation, and the optimal allocation is obtained to obtain the correlation defect source cost dispersion, which provides a reliable input for subsequent insulation state evaluation.

[0043] Preferably, in an embodiment of the present application, please refer to Figure 2 which shows a flow chart of a method for obtaining a cross-period defect source correlation cost matrix provided by an embodiment of the present application, specifically comprising: Step S201: In each monitoring period and the adjacent last monitoring period, respectively extract one event to form an event binary tuple.

[0044] Firstly, the event binary tuple is extracted and constructed for cross-period event comparison, so as to track the evolution trajectory of the defect source.

[0045] Step S202: According to the data difference of each dimension of the two discharge characteristic vectors, the cost component of each dimension of the corresponding event binary tuple is obtained.

[0046] Since the discharge characteristic vector is composed of multiple dimensions, each dimension represents different physical characteristics, so according to the data difference of each dimension of the two discharge characteristic vectors, the cost component of each dimension of the corresponding event binary tuple is obtained, which avoids the information loss caused by relying on a single indicator only, and improves the comprehensiveness of identification.

[0047] As an example, in the event pair, the discharge time centroid of the event of the monitoring period with the largest time sequence is subtracted from the discharge time centroid of the event of the monitoring period with the smallest time sequence, and the difference is taken as the cost component of the corresponding dimension, which makes the advance of the time centroid produce a negative value cost, which plays a rewarding role in the subsequent optimal allocation, so as to embed the degradation physical law of the core into the correlation model; And the other three cost components are based on the fact that the characteristics of the same defect source should remain relatively stable in a short time. The absolute values of the differences of the two events in the dimensions of discharge energy, spectral centroid and response waveform entropy are taken as the cost components of the corresponding dimensions, respectively, to show the same-dimensional data difference of the discharge feature vector through the cost components.

[0048] Step S203: All cost components in each dimension are combined into a cost component matrix, and a cross-period defect source correlation cost matrix is obtained after linear fusion with a preset weight.

[0049] Further, all event pairs in each dimension are collected and sorted, providing a basis for subsequent optimal allocation and avoiding local comparison distortion.

[0050] As an example, the serial number of the event in the monitoring period to which it belongs is used to form a two-dimensional coordinate of the matrix, and the corresponding relationship between the elements in each matrix and the event pair is unified. Since the above four original cost components have different dimensions and value ranges, in order to combine them in a unified scale, Z-score standardization is performed on each type of cost component before fusion, that is, based on the mean and standard deviation of the element values in the cost component matrix in each dimension, standardization processing is performed.

[0051] The preset weight adopts a time offset priority strategy, that is, the weight is [0.5, 0.15, 0.2, 0.15], and the cost components in the four dimensions of discharge time centroid, discharge energy, spectral centroid and response waveform entropy are linearly weighted and fused in order. The element values in the same position in the four matrices are weighted and fused to obtain a cross-period defect source correlation cost matrix corresponding to each monitoring period.

[0052] It should be noted that the analysis process of each event pair is consistent, and the analysis process of the cross-period defect source correlation cost matrix of each monitoring period is consistent. Only one example is described here, and the first monitoring period in the time domain is skipped. In other embodiments of the present application, the implementer can also use other weight setting methods such as equal weight [0.25, 0.25, 0.25, 0.25].

[0053] Preferably, in one embodiment of the present application, considering that the Jonker-Volgenant algorithm is a high-efficiency optimal allocation algorithm, which can find the global optimal matching in the cost matrix, therefore, based on the Jonker-Volgenant algorithm, the optimal allocation is performed on the cross-cycle defect source correlation cost matrix, and the discrete event identification problem is converted into a path tracking problem with clear physical meaning; The optimal matching outputs a set of successfully matched event binary tuples, representing the tracked defect source evolution path; and outputs a set of correlated defect source generation values, containing the final generation value corresponding to each successfully matched path, which can be used as a key data basis for individual behavior alienation degree; Considering that the dispersion degree of the final generation value reflects the abnormality degree of the trajectory, the greater the dispersion degree, the more significant the difference between the evolution trajectory and the group, which often means that the defect source has shown a deterioration trend of insulation behavior, therefore, according to the dispersion characteristics of the elements in the set of correlated defect source generation values, the correlation defect source cost dispersion degree is obtained, representing the degree of insulation deterioration.

[0054] As an example, the dispersion characteristics of the elements are represented in the form of variance, and the variance of the elements in the set of correlated defect source generation values is used as the correlation defect source cost dispersion degree.

[0055] It should be noted that the Jonker-Volgenant algorithm is prior art, in other embodiments of the present application, the implementer can use other optimal allocation algorithms such as the Auction Algorithm, or use the coefficient of variation instead of the variance to calculate the correlation defect source cost dispersion degree.

[0056] When the insulation state changes from global slow aging dominated by a large number of random micro-discharges with diverse characteristics to a dangerous state dominated by a high-risk local defect with highly convergent discharge behavior, the distribution density and dispersion degree of the discharge feature vector of a single discharge event in the multi-dimensional feature space will change fundamentally. In this step, the discharge system structure entropy is obtained according to the distribution of the discharge feature vector in each monitoring period, which can convert the dispersed single discharge feature vector into an overall order index, intuitively reflecting the transition process of the insulation system from the disordered state dominated by random micro-discharges to the ordered state dominated by local defects, and more easily capturing the early warning of insulation degradation and enhancing the early warning ability.

[0057] Preferably, in one embodiment of the present application, considering that a fixed similarity judgment standard cannot adapt to the dynamic changes of the system state, therefore, an adaptive scale benchmark is first calculated, which can represent the overall dispersion degree of all discharge behaviors in each monitoring period; Specifically, the Euclidean distance between the discharge feature vectors of any two events in each monitoring period is obtained, and the period discharge dispersion is obtained according to the discrete characteristics of all Euclidean distances, representing the diversity of all discharge event patterns in the monitoring period.

[0058] As an example, the standard deviation of the Euclidean distance between all discharge feature vectors in the monitoring period is taken as the period discharge dispersion, which is expressed in the form of standard deviation.

[0059] Each discharge event in the monitoring period is regarded as a node of a graph, and the relationship between events can be described in graph theory. Considering that the dispersion degree of discharge patterns in different monitoring periods is quite different, directly using a fixed threshold to judge the "near or far" of the Euclidean distance has poor dynamic adaptability, so the period discharge dispersion is introduced as a scale reference. Based on the period discharge dispersion as the kernel width, the Euclidean distance is mapped by a Gaussian kernel function to obtain the edge weight between nodes, i.e. similarity, and the similarity between each pair of events is filled into a The matrix, N is the total number of events in the analyzed monitoring period, and the event is completely similar to itself, with a similarity of 1. The discharge event similarity matrix is constructed, ensuring that the determination criteria of "proximity" or "similarity" can be dynamically adjusted according to the system state, thereby ensuring the robustness of the analysis.

[0060] The discharge event similarity matrix is regarded as an adjacency matrix of a graph, and a normalized graph Laplacian matrix is calculated therefrom. Then, the graph Laplacian matrix is subjected to eigenvalue decomposition, and the Shannon entropy of the eigenvalue is taken as the discharge system structure entropy.

[0061] The discharge system structure entropy is a nonlinear index for measuring the complexity of the entire discharge system structure. High entropy corresponds to a global, diversified and random discharge state, while low entropy clearly points to a dangerous state where one or a few discharge modes dominate and are highly ordered. In order to accurately quantify the structural transition from disorder (high entropy) to order (low entropy) within the entire discharge system.

[0062] It should be noted that the Euclidean distance between vectors, the Gaussian kernel function, and the process of calculating the normalized graph Laplacian matrix and performing eigenvalue decomposition using graph theory are well-known techniques and will not be described in detail.

[0063] Considering that the change in the structural entropy of the discharge system reflects the evolution trend of the overall orderliness and complexity of the insulation system, and that the dispersion of the associated defect source cost reflects the stability difference of different defect sources in cross-cycle matching, exhibiting individual deviation characteristics, an insulation state diagnostic space is constructed based on the change in the structural entropy of the discharge system in each monitoring cycle compared to the previous monitoring cycle, combined with the dispersion of the associated defect source cost. This allows for the simultaneous quantification of the global state of the system and local abnormal deviations, thus forming a two-dimensional diagnostic space and achieving a more comprehensive state diagnosis.

[0064] Preferably, in one embodiment of the present invention, the change in the structural entropy of the discharge system in each monitoring cycle compared to the previous monitoring cycle is represented by taking the difference. The difference between the structural entropy of the discharge system in each monitoring cycle and the structural entropy of the discharge system in the adjacent previous monitoring cycle is used as the abscissa of the diagnostic point in each monitoring cycle. The positive direction of the horizontal axis represents the decrease in entropy, which corresponds to the system moving from disorder to order and is the direction of structural deterioration.

[0065] The dispersion of associated defect source cost is used as the ordinate of the diagnostic point in each monitoring cycle. Corresponding to the vertical axis, the positive direction of the vertical axis represents the appearance of an "abnormal" defect source in the system whose evolutionary behavior is far different from that of the population, which is the direction of behavioral degradation. An insulation condition diagnostic space is constructed using diagnostic points from all monitoring cycles.

[0066] Step S3: Perform multi-level condition assessment based on the distribution of diagnostic points in the insulation condition diagnostic space.

[0067] In the insulation condition diagnosis space, the coordinate distribution of each diagnosis point reflects the insulation condition of the system. Therefore, multi-level condition assessment is performed based on the distribution of diagnosis points in the insulation condition diagnosis space.

[0068] Preferably, in one embodiment of the present invention, please refer to Figure 3 The diagram illustrates a flowchart of a method for multi-level state assessment provided by an embodiment of the present invention, specifically including: Step S301: When the diagnostic point of the monitoring cycle is in the preset target quadrant of the insulation state diagnostic space, mark the corresponding diagnostic point as the state evaluation point, obtain the square of the distance from the state evaluation point to the origin as the degradation severity index, and obtain the angle of the state evaluation point in the preset target quadrant as the degradation mode angle.

[0069] Considering that in a healthy state, the diagnostic point should fluctuate randomly around the origin, when both the system structure and individual behavior show signs of deterioration, it indicates that the insulation state may be abnormal. Therefore, the first quadrant is set as the preset target quadrant. When the horizontal and vertical coordinates of the diagnostic point are both greater than 0, it is marked as a state assessment point. The severity of the insulation state deviating from the normal range is quantified by considering the square of the distance from the state evaluation point to the origin, so it is taken as the deterioration severity index; The angle of the state evaluation point reveals the dominant driving mode of the deterioration process, providing a deeper insight for fault diagnosis, so the angle of the state evaluation point in the preset target quadrant is obtained as the deterioration mode angle.

[0070] As an example, the ratio of the ordinate to the abscissa is taken as the independent variable, and after mapping through the arctan(x) function, the deterioration mode angle is obtained.

[0071] Step S302: According to the distribution of the deterioration severity index of the state evaluation point, combined with the deterioration mode angle, multi-level state evaluation is carried out.

[0072] The numerical distribution of the deterioration severity index of the state evaluation point can depict the degree of the insulation state deviating from the normal range, so multi-level state evaluation is carried out based on the distribution of the deterioration severity index, avoiding simple binary (normal / fault) evaluation of the health state of the insulation system.

[0073] As an example, first, the space is regionally divided to establish a quantitative state evaluation benchmark. At the initial stage of device operation or known health state, the system continuously calculates the deterioration severity index. Since the random discharge activity of the insulation system in the healthy state will cause the diagnostic point to fluctuate in a small range near the origin, a background deterioration severity index is generated. Based on statistical methods, the health threshold is obtained based on the deterioration severity index of all state evaluation points; Specifically, the 99% quantile of the deterioration severity index is obtained as the health threshold, providing a basis for grading determination. To capture the trajectory of the state evaluation point, the state evaluation points in the current preset time domain neighborhood are analyzed, and the nearest 6 historical monitoring periods and the current monitoring period in the time domain are taken as the current preset time domain neighborhood. If the deterioration severity index of the state evaluation points in the current preset time domain neighborhood is less than the health threshold, it means that the trajectory of the state evaluation point is constrained in the "healthy / stable region" defined by the health threshold, that is, the evolution of the individual behavior and the structural evolution of the system as a whole have not formed a synergistic amplification effect. This corresponds to the health of the device insulation state, or only the risk controllable, diffuse global slow aging, the system state is normal, no warning is generated, and the operation and maintenance personnel can maintain the regular inspection period; If there are both degradation severity indexes less than the health threshold and degradation severity indexes greater than or equal to the health threshold in the degradation severity indexes of the state evaluation points in the preset time domain neighborhood, it is indicated that the state evaluation points begin to appear intermittent and short-term crossing of the boundary of the "health / stable region", but fail to form a certain motion trajectory away from the origin, indicating that structural or behavioral abnormalities begin to appear in the system, but a stable and continuously developing degradation path has not yet been formed.

[0074] This may correspond to the initial stage of the germination of a certain potential defect, or an unstable discharge source activated under a certain operating condition. This is a key transition stage from health to failure, and a warning of concern is issued, suggesting that the operation and maintenance personnel appropriately increase the frequency of monitoring data analysis of the equipment, for example, from daily analysis to hourly analysis, so as to capture the subsequent evolution trend; If all the degradation severity indexes of the state evaluation points in the preset time domain neighborhood are greater than or equal to the health threshold, it is indicated that the trajectory of the state evaluation points has clearly deviated from the "health / stable region", meaning that the evolution of the system has entered an irreversible and self-reinforcing degradation channel. The abnormal evolution of individual defects and the ordering of the system structure form a positive feedback, which is a clear signal of the rapid development of a high-risk local defect and an important precursor before the occurrence of an inter-turn short circuit, and a degradation warning is issued.

[0075] In other embodiments of the present application, the health threshold and the setting of the preset time domain neighborhood can be adjusted by the implementer.

[0076] In an embodiment of the present application, after the degradation warning is issued, the degradation mode angle is further used to reveal the dominant driving mode of the degradation process, providing more basis for health management, specifically including: When the degradation mode angle of the latest state evaluation point is in the preset low-angle interval, it means that the trajectory is close to the horizontal axis, and the main driving force is the change value of the discharge system structural entropy. Corresponding to the situation inside the reactor, an existing structural defect (such as an inter-turn insulation interface crack) is rapidly expanding, and its discharge characteristics gradually dominate the entire system, causing the discharge mode of the system to quickly change from diversification (high entropy) to singleness (low entropy), so it is determined that a single structural defect is rapidly developing, and it is recommended that the operation and maintenance personnel prioritize the inspection of areas with stress concentration or manufacturing process defects inside the reactor; When the degradation mode angle of the latest state evaluation point is in the preset high-angle interval, it is indicated that the trajectory is close to the vertical axis, and the main driving force is the correlation defect source cost dispersion. Corresponding to the situation inside the reactor, the insulating medium at multiple positions simultaneously enters an unstable state, and its discharge behavior changes dramatically and inconsistently, but has not yet formed an absolutely dominant defect source. This phenomenon is usually related to the overall performance degradation of the insulating material, so it is determined that it is a multi-point diffuse degradation, and it is recommended that the operation and maintenance personnel perform a comprehensive insulating state evaluation test on the equipment to confirm the overall insulating health level. When the deterioration mode angle of the latest state evaluation point is in the preset medium angle interval, it means that the two processes of structure ordering and behavior alienation are coordinated and advanced. This indicates that a dominant defect has been formed and is rapidly evolving itself, which is the highest risk signal before the occurrence of inter-turn short circuit, so it is determined that the mature dominant defect is coordinated and evolved, and it is suggested that the device is taken out of operation and arranged for shutdown maintenance.

[0077] As an example, the preset low angle interval is (0, 30], the preset medium angle interval is (30, 60], and the preset high angle interval is (60, 90], in degrees; in other embodiments of the application, the implementer can adjust it by himself.

[0078] Finally, a report containing a clear diagnosis conclusion and targeted operation and maintenance suggestions is generated.

[0079] In summary, in order to solve the technical problems that the prior art is difficult to effectively extract early insulation degradation information under strong transient overvoltage interference, and cannot accurately predict faults and health management, the application provides a dry-type air-core reactor inter-turn fault identification method with optimized data acquisition. The application first acquires the discharge feature vector of each event; further acquires the associated defect source cost dispersion according to the same dimension data difference of the discharge feature vectors between adjacent monitoring periods; further acquires the discharge system structure entropy according to the distribution of the discharge feature vectors in each monitoring period; further constructs an insulation state diagnosis space by combining the change of the discharge system structure entropy of adjacent monitoring periods and the associated defect source cost dispersion; and finally performs multi-level state evaluation according to the distribution of the diagnosis points in the insulation state diagnosis space. By constructing the defect source associated cost dispersion through cross-period discharge feature difference, combining the change of the discharge system structure entropy to form the insulation state diagnosis space, and performing multi-level state evaluation based on the distribution of the diagnosis points, early degradation information is captured, the severity of insulation degradation and the dominant mode are accurately predicted and quantified, and the abnormal evolution process is provided as a reliable basis for health management.

[0080] It should be noted that the above-mentioned embodiments of the application are in the order of description only, and do not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or may be advantageous.

[0081] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment mainly describes the difference from other embodiments.

Claims

1. A method for identifying inter-turn faults in dry-type air-core reactors based on optimized data acquisition, characterized in that, The method includes: Obtain the discharge feature vector of each event recorded in each preset monitoring period; Based on the same-dimensional data difference of the discharge feature vector between each monitoring cycle and the adjacent monitoring cycle, a cross-cycle defect source association cost matrix is ​​constructed, and the associated defect source cost dispersion is obtained by optimal allocation; based on the distribution of the discharge feature vector within each monitoring cycle, the discharge system structure entropy is obtained; based on the change of the discharge system structure entropy in each monitoring cycle compared to the previous monitoring cycle, combined with the associated defect source cost dispersion, an insulation state diagnostic space is constructed. Multi-level condition assessment is performed based on the distribution of diagnostic points in the insulation condition diagnostic space.

2. The method for identifying inter-turn faults in a dry-type air-core reactor based on optimized data acquisition according to claim 1, characterized in that, The discharge feature vector includes: the discharge time centroid, discharge energy, spectral centroid, and response waveform entropy of the voltage response sequence for each event.

3. The method for identifying inter-turn faults in a dry-type air-core reactor based on optimized data acquisition according to claim 2, characterized in that, The method for obtaining the cross-cycle defect source correlation cost matrix includes: In each monitoring cycle and the adjacent previous monitoring cycle, one event is extracted to form an event tuple; Based on the data difference of each dimension of the two discharge feature vectors, the cost component of each dimension of the corresponding event tuple is obtained; all the cost components of each dimension are used to form a cost component matrix, and after linear fusion with preset weights, a cross-cycle defect source association cost matrix is ​​obtained.

4. The method for identifying inter-turn faults in a dry-type air-core reactor based on optimized data acquisition, as described in claim 3, is characterized in that... The method for obtaining the cost component includes: In the event tuple, the difference between the discharge time centroid of the event with the longest time sequence and the discharge time centroid of the event with the shortest time sequence in the monitoring period is used as the cost component of the corresponding dimension; the absolute values ​​of the differences between the two events in the dimensions of discharge energy, spectrum centroid and response waveform entropy are used as the cost components of the corresponding dimensions.

5. The method for identifying inter-turn faults in a dry-type air-core reactor based on optimized data acquisition according to claim 3, characterized in that, The method for obtaining the discreteness of the associated defect source cost includes: The Jonker-Volgenant algorithm is used to optimally allocate the cross-cycle defect source association cost matrix to obtain the associated defect source cost set. Based on the discrete characteristics of the elements in the associated defect source cost set, the cost dispersion of the associated defect sources is obtained.

6. The method for identifying inter-turn faults in a dry-type air-core reactor based on optimized data acquisition according to claim 1, characterized in that, The method for obtaining the structural entropy of the discharge system includes: Obtain the Euclidean distance between the discharge feature vectors of any two events within each monitoring period; obtain the periodic discharge dispersion based on the discrete characteristics of all Euclidean distances; and obtain the discharge event similarity matrix by mapping the Euclidean distance using the periodic discharge dispersion as the kernel width and a Gaussian kernel function. Obtain the graph Laplacian matrix of the discharge event similarity matrix and perform eigenvalue decomposition to obtain the Shannon entropy of the eigenvalues ​​as the structural entropy of the discharge system.

7. The method for identifying inter-turn faults in a dry-type air-core reactor based on optimized data acquisition according to claim 1, characterized in that, The method for constructing the insulation condition diagnostic space includes: The difference between the discharge system structure entropy of each monitoring cycle and the discharge system structure entropy of the adjacent previous monitoring cycle is used as the abscissa of the diagnostic point of each monitoring cycle; the dispersion of the associated defect source cost is used as the ordinate of the diagnostic point of each monitoring cycle. An insulation condition diagnostic space is constructed using diagnostic points from all the monitoring cycles.

8. The method for identifying inter-turn faults in a dry-type air-core reactor based on optimized data acquisition according to claim 1, characterized in that, The method for performing multi-level state assessment includes: When the diagnostic point of the monitoring cycle is in the preset target quadrant of the insulation state diagnostic space, the corresponding diagnostic point is marked as a state evaluation point, and the square of the distance from the state evaluation point to the origin is obtained as the degradation severity index; and the angle of the state evaluation point in the preset target quadrant is obtained as the degradation mode angle. Based on the distribution of the degradation severity index at the state assessment points, a multi-level state assessment is performed in conjunction with the degradation mode angle.

9. The method for identifying inter-turn faults in a dry-type air-core reactor based on optimized data acquisition according to claim 8, characterized in that, The method for multi-level state assessment based on the distribution of the degradation severity index at the state assessment points and in conjunction with the degradation mode angle includes: Based on the obtained health threshold of the degradation severity index of all the state evaluation points, if the degradation severity index of all the state evaluation points in the current preset time domain neighborhood is less than the health threshold, the system is determined to be in normal condition. If, within the current preset time-domain neighborhood, there are simultaneously two severe degradation indices for the state evaluation points that are both less than the health threshold and greater than or equal to the health threshold, a warning of concern is issued. If the severity index of the degradation of the state evaluation points in the current preset time domain is greater than or equal to the health threshold, a degradation warning is issued.

10. The method for identifying inter-turn faults in a dry-type air-core reactor based on optimized data acquisition according to claim 9, characterized in that, Following the issuance of the degradation warning, the following also includes: When the degradation mode angle of the latest state assessment point is in a preset low angle range, it is determined that a single structural defect is rapidly developing. When the degradation mode angle of the latest state evaluation point is in the preset high angle range, it is determined to be multi-point diffuse degradation. When the degradation mode angle of the latest state assessment point is within the preset medium angle range, it is determined to be a mature dominant defect co-evolution.

Images